[prev] [up] [next]


Smalltalk systems provide an integrated programming environment, which is quite different to almost any other programming system. The difference is that in Smalltalk, both your program and the IDE live in and are part of the same object-world. In Smalltalk, you are not editing a file or code from which an application is built. Instead, you are adding code to the IDE to add your application(s) to it. This does not mean, that it is not possible or common to create a stand alone application for deployment eventually. But during the development, your code and the IDE are one program. Even more striking is the fact, that often Smalltalk programmers have ALL of their code they ever developed in their IDE. This makes reuse much more straight forward.

The Smalltalk IDE is a really integrated environment, not just a bunch of editing tools wrapped into as single frame.

This programming environment allows up-to date access to an object's instance variables, to a classes definition, the set of messages which are implemented there (i.e. the protocol), the source code of those methods, the execution state and much more.
In addition, cross references can be searched (who sends which message, who implements which message, who references a variable etc.), breakpoints can be set and modifications can be made to a class.
Modifications to a class can be made at any time - even while there are active instances of it around; any modification is immediately effective (that does make a difference in programmers productivity !).

Defining new Classes

In Smalltalk, writing programs consists of defining classes (whose instances represent the objects of your application), and defining some protocol (i.e. methods, which define how those objects react to messages).

Every Smalltalk system also includes a (usually) large collection of useful classes, which provide building blocks and cover a wide area of typical objects/operations. Therefore, a big part of learning how to write Smalltalk programs consists of learning how to use the class library.

Although it is possible to define new classes programatically (i.e. by sending definition-messages to some object in a workspace or by a program), classes are normally defined using a browser. This tool provides functions for most programming tasks and will be your most-used tool when working with Smalltalk.
Therefore, take some time and play around in a System browser to get a feeling for it.
Also, have a look at the browser's documentation and the new improved browser's documentation.

You can open a System browser either via the launcher's menu, via an inspector's menu, or by sending a "browse"-message to any object.

All classes are organized by various aspects: inheritance hierarchy, package-ownership and class category. When opened, the default organization shown in the browser is "by class category".
Class Categories exist only for organization - to allow grouping of classes with similar behavior. For example, all container classes are organized under categories starting with "collection". This makes it easy to find a class.

To define a new class, first add a new class category using the class-category-lists popup menu:

[fig: adding category1]

enter the name of the new category:

[fig: adding category2]

and click on the 'create' button of the entry box.

The browser will add this new category to its list and select (i.e. highlight) it:

[fig: adding category2]

then, select the class-lists new class menu item,

[fig: adding class]

to get a definition template into the codeView:

[fig: template1]

Fill out the template

[fig: template2]

and accept it using codeViews popup menu.

[fig: accepting]

This creates your new class and the browser will update its class-list view, showing your new class:

[fig: accepted]

In this example, we defined a Person-class, and defined two instance variables named name and firstname. I.e. instances of this class will be able to hold references to two other objects, and those will be referred to as name and firstname.

Creating an Instance of a Class

Now, our new class is known to the system, and new instances can be created by sending the new message to the class:
    Person new
You can type in the above expression into any code view or workspace, select it and use one of the "DoIt" / "PrintIt" / "InspectIt" menu functions.

You don't have to compile, build or otherwise care for the compilation details. The class is ready to be used at the moment you defined and accepted its definition.

You may wonder, how our new class 'knows' how to respond to the "new"-message - the answer is that all classes are themself instances of a so called metaclass, which itself inherits from the Behavior class.
Behavior defines quite a bunch of useful protocol, one being the "new"-method.
If you are curious, have a look at the Behavior class in your browser and see what other useful protocol is found there.

Defining Methods

Inherited Protocol

So far, instances of the Person class do not provide any special protocol - however, a lot is inherited from its superclass: Object.

Therefore, we can already ask instances about their class or class-name, we can compare instances, ask an instance for a copy of itself and ask it for its printed string representation.
Try a few expressions (using printIt) in a workspace:

Its printed representation (called printString):

     Person new printString
Display it on the Transcript:
     Transcript show:(Person new).
     Transcript cr.
another method to display it on the Transcript:
     Person new printOn:Transcript.
Comparing for identity:
     |p1 p2|

     p1 := Person new.
     p2 := Person new.
     p1 == p2
Asking for a copy:
     |p1 p2|

     p1 := Person new.
     p2 := p1 copy.
     Transcript showCR:p1.
     Transcript showCR:p2.
Also, have a look at the implementation of some of the above protocol - especially, have a look into the "printString" implementation and understand that there is another (more basic) functionality involved (to give you a hint: figure out what the role of the "printOn:" method is ...)

In order to do someting useful with persons, we must tell them how to respond to more specific messages - i.e. we have to add some more protocol. Also, we might want to redefine some of the inherited protocol, for example, we may want to change the way instances represent themself in their printString.

The task of adding protocol (i.e. methods) is much like what we did to define the class - first we have to define a method category by selecting the new method category item from the method-category-list:

[fig: new method category]

enter some descriptive name:

[fig: new method category]

and click the 'create' button.

Then, either ask for a methods template code or simply type in the code and accept it via the codeViews popup menu.

Adding Protocol for Access

As a first step, we need some protocol to change our person object's contents - i.e. to access the name and firstName.
Lets start with a name-setting method:

[fig: new method category]

The above method defines how instances of our class will react to the setName: keyword message - a message with one argument.
Here, the "setName:" method simply binds the instance variable called name to whatever was given as the messages argument.

We can check this in a workspace, by evaluating:

    Person new setName:'Miller'
Of course, this is pretty useless, since we should have a look into our person instance, in order to validate that it really did what it is supposed to ...
... for this, Smalltalk provides a tool called inspector, which allows us to peek into an object.
Try the above expression again, this time using the inspectIt menu function (instead of doIt). This evaluates the expression and opens an inspector on the result:

[fig: inspector on a person]

click on the name item in the list, to see its value:

[fig: inspector on a person]

Also, open an inspector on a fresh person - and see what the instance variables are initialized with (automatically).

In addition to setting a persons name, we would also like to ask a person for its name.
Similar to the above setter-method, we need a corresponding getter-method, which returns some information about the person object.

Returning a Value from a Method

To return some value from a message, a return-expression is used - its syntax is:
     ^ expression
i.e. a little up-arrow (^) followed by an expression.

this evaluates the expression and returns the expressions value from the method - effectively returning that value as the value of the message send (remember: all methods are invoked due to some message-expression evaluation).

Now, that we know that every message-send evaluates to something, and all messages are actually implemented in some method, you may ask what value is returned if no explicit return expression is evaluated in a method (as in our first example method).
The answer is:
if no explicit return value is given in a method, the value returned is the original receiver of the message

This makes sense, because if the returned value is the original object, it may be the target of another message send, as in:

    (Person new setName:'Miller') setFirstName:'Fred'
or, if we know the preceedence rules:
    Person new setName:'Miller' setFirstName:'Fred'
Many methods (and therefore: many messages) return the original receiver - however, there are some noticable exceptions, which form a trap to beginners. If you are in doubt, better do not depend on a methods returnValue and write things down explicit:

    newPerson := Person new.
    newPerson setName:'Miller'.
    newPerson setFirstName:'Fred'.
Although it takes more to type this, it definitely makes things clearer and less error prone.

Now, we are ready to add the getter-method for the persons name;
add the following method to the Person class:

	"return the persons name"

	^ name
As an excercise, add corresponding set- and get- methods for the firstname instance variable.

Redefining the "printOn:"-Method

Up to now, persons did not show themself nicely when displayed in the Transcript, or when printed via printIt.
In order to change this, we need some background information:

Whenever an object is to be displayed somewhere (for example: in the Transcript), the object in question is asked for its printString (using the printString -message).
Therefore, it would make sense, to redefine this, to get a nicer printout of persons.
However, if you had a look at the implementation of this message, you might have noticed, that the printStirng implementation simply forwards the real work to some other method, by sending the printOn: message.
This one is supposed to append a printed representation to some stream, which gets passed as argument.
The reason for this is to avoid the construction of many temporary strings, when objects are printed on the printer or a printed representation is stored into a file; in those cases, the printOn: message is sent directly, passing the destination stream as argument.

So, we better redefine the printOn: method, and a possible implementation could be:

	"send a printed representation of myself
	 to aStream."

	aStream nextPutAll:'Person('.
	aStream nextPutAll:'name='.
	name printOn:aStream.
	aStream nextPutAll:' firstname='.
	firstname printOn:aStream.
	aStream nextPutAll:')'.
An interesting detail of the above code is that the name and firstname instance variables are themself asked to print on the stream, instead of passing them as argument of a "nextPutAll:" message.
This has the advantage that anything can be used as name/firstname; as opposed to "aStream nextPutAll:name" - which implies that its argument is some string-like object.

With that addition, our person object prints much nicer; try it with 'Fred Miller':


    p := Person new.
    p setName:'Miller'.
    p setFirstName:'Fred'.

    Transcript showCR:p

Changing a Method

Very often, an existing method has to be changed - either to fix a bug, to add functionality, or because you do not like its behavior.
To change some existing method, select it in the browser, make any changes to the text as required, and accept it again.

Lets change the printOn: method, to display persons in another format:

	"send a printed representation of myself
	 to aStream."

	aStream nextPutAll:'a person named '.
	firstname printOn:aStream.
	aStream space.
	name printOn:aStream.
Notice, that when you accept something in the browser, the code is compiled and installed as specified in the text - regardless of which method the text originally came from.
This allows you to take any other methods code as a template to start with.
I.e. you don't have to type in all of the methods text if you find some similar method.

Adding more Instance Variables

In much the same way, more instance variables can be added to the person-class.
Click on the class, to get its definition, add another identifier to the instanceVariableNames section, and accept.

For the curious:
You should notice, that since you changed the definition of person, any existing person object cannot be an instance of this 'new' class, because they do neither know about, nor hold on any of the additional instance variables.
Therefore, any existing instances are now made instances of a so called obsolete person class - the old person class will stay around in the system until no more obsolete person objects are in the system.

As an exersize, add an instance variable called telno and also add corresponding access protocol.
By now, you may wonder if such mechanic things as defining access methods could not be done automatically - you are right !
The browser's method category menu provides a function called "generate access methods". This defines get- and set-methods for either a single or all instance variables - the methods will be named according to the instance variables name(s).

To try it, select a name from the instance variable list and invoke that menu item:

[fig: new method category]

Redefining the class protocol

As mentioned earlier, every class is an instance of its metaclass - therefore, the classes protocol (i.e. messages such as #new) are defined in its metaclass.
We can change and/or enhance a classes class-protocol in the same way we do to the instance protocol; in the browser, the class/instance toggle is used to switch between the two protocol sets.

As an example, let us redefine our persons #new method, to automatically initalize the name fields of the newly created person object:

then, add the following method:

	newPerson := super new.
	newPerson setName:'*** no name yet ***'.
	^ newPerson.
The interesting thing here is the "super new". This invokes the original #new method, as inherited from some superclass (i.e. the one we have overwritten).
Since we have redefined the #new method, a simple "self new" would lead to an endless recursion.

Continue in "Creating a Web Application".

Copyright © 1996 Claus Gittinger Development & Consulting
Copyright © eXept Software AG

<cg at exept.de>

Doc $Revision: 1.9 $ $Date: 1998/10/30 21:02:51 $